[llvm] [NFC][AMDGPU] Auto generate check lines for some codegen tests (PR #137534)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Apr 27 10:12:35 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Shilei Tian (shiltian)
<details>
<summary>Changes</summary>
Make preparation for #<!-- -->137488.
---
Patch is 314.42 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/137534.diff
11 Files Affected:
- (modified) llvm/test/CodeGen/AMDGPU/dag-divergence.ll (+69-18)
- (modified) llvm/test/CodeGen/AMDGPU/flat-offset-bug.ll (+279-26)
- (modified) llvm/test/CodeGen/AMDGPU/lds-misaligned-bug.ll (+857-73)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.dec.ll (+1507-187)
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.inc.ll (+1570-166)
- (modified) llvm/test/CodeGen/AMDGPU/load-select-ptr.ll (+93-35)
- (modified) llvm/test/CodeGen/AMDGPU/memory-legalizer-store-infinite-loop.ll (+20-5)
- (modified) llvm/test/CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll (+469-125)
- (modified) llvm/test/CodeGen/AMDGPU/waitcnt-vscnt.ll (+392-74)
- (modified) llvm/test/Transforms/InferAddressSpaces/AMDGPU/basic.ll (+119-62)
- (modified) llvm/test/Transforms/InferAddressSpaces/AMDGPU/mem-intrinsics.ll (+88-31)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/dag-divergence.ll b/llvm/test/CodeGen/AMDGPU/dag-divergence.ll
index 9f83393d88061..cdf4a88814dfc 100644
--- a/llvm/test/CodeGen/AMDGPU/dag-divergence.ll
+++ b/llvm/test/CodeGen/AMDGPU/dag-divergence.ll
@@ -1,11 +1,29 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
-; GCN-LABEL: {{^}}private_load_maybe_divergent:
-; GCN: buffer_load_dword
-; GCN-NOT: s_load_dword s
-; GCN: flat_load_dword
-; GCN-NOT: s_load_dword s
define amdgpu_kernel void @private_load_maybe_divergent(ptr addrspace(4) %k, ptr %flat) {
+; GCN-LABEL: private_load_maybe_divergent:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_add_i32 s12, s12, s17
+; GCN-NEXT: s_mov_b64 s[22:23], s[2:3]
+; GCN-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8
+; GCN-NEXT: s_mov_b64 s[20:21], s[0:1]
+; GCN-NEXT: s_add_u32 s20, s20, s17
+; GCN-NEXT: s_addc_u32 s21, s21, 0
+; GCN-NEXT: buffer_load_dword v0, v0, s[20:23], 0 offen glc
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_load_dwordx2 s[0:1], s[8:9], 0x0
+; GCN-NEXT: s_mov_b32 flat_scratch_lo, s13
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: v_mov_b32_e32 v2, s1
+; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0
+; GCN-NEXT: v_lshlrev_b64 v[0:1], 2, v[0:1]
+; GCN-NEXT: v_add_u32_e32 v0, vcc, s0, v0
+; GCN-NEXT: v_addc_u32_e32 v1, vcc, v2, v1, vcc
+; GCN-NEXT: flat_load_dword v0, v[0:1]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: flat_store_dword v[0:1], v0
+; GCN-NEXT: s_endpgm
%load = load volatile i32, ptr addrspace(5) poison, align 4
%gep = getelementptr inbounds i32, ptr addrspace(4) %k, i32 %load
%maybe.not.uniform.load = load i32, ptr addrspace(4) %gep, align 4
@@ -13,15 +31,27 @@ define amdgpu_kernel void @private_load_maybe_divergent(ptr addrspace(4) %k, ptr
ret void
}
-; GCN-LABEL: {{^}}flat_load_maybe_divergent:
-; GCN: s_load_dwordx4
-; GCN-NOT: s_load
-; GCN: flat_load_dword
-; GCN-NOT: s_load
-; GCN: flat_load_dword
-; GCN-NOT: s_load
-; GCN: flat_store_dword
define amdgpu_kernel void @flat_load_maybe_divergent(ptr addrspace(4) %k, ptr %flat) {
+; GCN-LABEL: flat_load_maybe_divergent:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0
+; GCN-NEXT: s_add_i32 s12, s12, s17
+; GCN-NEXT: s_mov_b32 flat_scratch_lo, s13
+; GCN-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: v_mov_b32_e32 v0, s2
+; GCN-NEXT: v_mov_b32_e32 v1, s3
+; GCN-NEXT: flat_load_dword v0, v[0:1]
+; GCN-NEXT: v_mov_b32_e32 v2, s1
+; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0
+; GCN-NEXT: v_lshlrev_b64 v[0:1], 2, v[0:1]
+; GCN-NEXT: v_add_u32_e32 v0, vcc, s0, v0
+; GCN-NEXT: v_addc_u32_e32 v1, vcc, v2, v1, vcc
+; GCN-NEXT: flat_load_dword v0, v[0:1]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: flat_store_dword v[0:1], v0
+; GCN-NEXT: s_endpgm
%load = load i32, ptr %flat, align 4
%gep = getelementptr inbounds i32, ptr addrspace(4) %k, i32 %load
%maybe.not.uniform.load = load i32, ptr addrspace(4) %gep, align 4
@@ -34,12 +64,33 @@ define amdgpu_kernel void @flat_load_maybe_divergent(ptr addrspace(4) %k, ptr %f
; last values are divergent due to the carry in glue (such that
; divergence needs to propagate through glue if there are any non-void
; outputs)
-; GCN-LABEL: {{^}}wide_carry_divergence_error:
-; GCN: v_sub_u32_e32
-; GCN: v_subb_u32_e32
-; GCN: v_subb_u32_e32
-; GCN: v_subb_u32_e32
define <2 x i128> @wide_carry_divergence_error(i128 %arg) {
+; GCN-LABEL: wide_carry_divergence_error:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_ffbh_u32_e32 v0, v0
+; GCN-NEXT: v_ffbh_u32_e32 v4, v2
+; GCN-NEXT: v_add_u32_e64 v0, s[4:5], v0, 32 clamp
+; GCN-NEXT: v_ffbh_u32_e32 v1, v1
+; GCN-NEXT: v_add_u32_e32 v4, vcc, 32, v4
+; GCN-NEXT: v_min3_u32 v0, v0, v1, 64
+; GCN-NEXT: v_add_u32_e32 v0, vcc, 64, v0
+; GCN-NEXT: v_ffbh_u32_e32 v5, v3
+; GCN-NEXT: v_addc_u32_e64 v1, s[4:5], 0, 0, vcc
+; GCN-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
+; GCN-NEXT: v_min_u32_e32 v4, v4, v5
+; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc
+; GCN-NEXT: v_sub_u32_e32 v0, vcc, 0, v0
+; GCN-NEXT: v_mov_b32_e32 v3, 0
+; GCN-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc
+; GCN-NEXT: v_subb_u32_e32 v2, vcc, 0, v3, vcc
+; GCN-NEXT: v_subb_u32_e32 v3, vcc, 0, v3, vcc
+; GCN-NEXT: v_mov_b32_e32 v4, 0
+; GCN-NEXT: v_mov_b32_e32 v5, 0
+; GCN-NEXT: v_mov_b32_e32 v6, 0
+; GCN-NEXT: v_mov_b32_e32 v7, 0
+; GCN-NEXT: s_setpc_b64 s[30:31]
%i = call i128 @llvm.ctlz.i128(i128 %arg, i1 false)
%i1 = sub i128 0, %i
%i2 = insertelement <2 x i128> zeroinitializer, i128 %i1, i64 0
diff --git a/llvm/test/CodeGen/AMDGPU/flat-offset-bug.ll b/llvm/test/CodeGen/AMDGPU/flat-offset-bug.ll
index 54343fa820cba..1732dd0521e5f 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-offset-bug.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-offset-bug.ll
@@ -1,13 +1,40 @@
-; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9_11 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9_11 %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX11 %s
-; GCN-LABEL: flat_inst_offset:
-; GFX9_11: flat_load_{{dword|b32}} v{{[0-9]+}}, v[{{[0-9:]+}}] offset:4
-; GFX9_11: flat_store_{{dword|b32}} v[{{[0-9:]+}}], v{{[0-9]+}} offset:4
-; GFX10: flat_load_dword v{{[0-9]+}}, v[{{[0-9:]+}}]{{$}}
-; GFX10: flat_store_dword v[{{[0-9:]+}}], v{{[0-9]+}}{{$}}
define void @flat_inst_offset(ptr nocapture %p) {
+; GFX9-LABEL: flat_inst_offset:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: flat_load_dword v2, v[0:1] offset:4
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_u32_e32 v2, 1, v2
+; GFX9-NEXT: flat_store_dword v[0:1], v2 offset:4
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: flat_inst_offset:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, v0, 4
+; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX10-NEXT: flat_load_dword v2, v[0:1]
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_nc_u32_e32 v2, 1, v2
+; GFX10-NEXT: flat_store_dword v[0:1], v2
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: flat_inst_offset:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: flat_load_b32 v2, v[0:1] offset:4
+; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_nc_u32_e32 v2, 1, v2
+; GFX11-NEXT: flat_store_b32 v[0:1], v2 offset:4
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr inbounds i32, ptr %p, i64 1
%load = load i32, ptr %gep, align 4
%inc = add nsw i32 %load, 1
@@ -15,10 +42,34 @@ define void @flat_inst_offset(ptr nocapture %p) {
ret void
}
-; GCN-LABEL: global_inst_offset:
-; GCN: global_load_{{dword|b32}} v{{[0-9]+}}, v[{{[0-9:]+}}], off offset:4
-; GCN: global_store_{{dword|b32}} v[{{[0-9:]+}}], v{{[0-9]+}}, off offset:4
define void @global_inst_offset(ptr addrspace(1) nocapture %p) {
+; GFX9-LABEL: global_inst_offset:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: global_load_dword v2, v[0:1], off offset:4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_u32_e32 v2, 1, v2
+; GFX9-NEXT: global_store_dword v[0:1], v2, off offset:4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: global_inst_offset:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: global_load_dword v2, v[0:1], off offset:4
+; GFX10-NEXT: s_waitcnt vmcnt(0)
+; GFX10-NEXT: v_add_nc_u32_e32 v2, 1, v2
+; GFX10-NEXT: global_store_dword v[0:1], v2, off offset:4
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: global_inst_offset:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: global_load_b32 v2, v[0:1], off offset:4
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: v_add_nc_u32_e32 v2, 1, v2
+; GFX11-NEXT: global_store_b32 v[0:1], v2, off offset:4
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr inbounds i32, ptr addrspace(1) %p, i64 1
%load = load i32, ptr addrspace(1) %gep, align 4
%inc = add nsw i32 %load, 1
@@ -26,10 +77,51 @@ define void @global_inst_offset(ptr addrspace(1) nocapture %p) {
ret void
}
-; GCN-LABEL: load_i16_lo:
-; GFX9_11: flat_load_{{short_d16|d16_b16}} v{{[0-9]+}}, v[{{[0-9:]+}}] offset:8{{$}}
-; GFX10: flat_load_short_d16 v{{[0-9]+}}, v[{{[0-9:]+}}]{{$}}
define amdgpu_kernel void @load_i16_lo(ptr %arg, ptr %out) {
+; GFX9-LABEL: load_i16_lo:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: v_mov_b32_e32 v2, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v0, s0
+; GFX9-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-NEXT: flat_load_short_d16 v2, v[0:1] offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_add_u16 v2, v2, v2
+; GFX9-NEXT: flat_store_dword v[0:1], v2
+; GFX9-NEXT: s_endpgm
+;
+; GFX10-LABEL: load_i16_lo:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX10-NEXT: v_mov_b32_e32 v2, 0
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s0, s0, 8
+; GFX10-NEXT: s_addc_u32 s1, s1, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, s0
+; GFX10-NEXT: v_mov_b32_e32 v1, s1
+; GFX10-NEXT: flat_load_short_d16 v2, v[0:1]
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_u16 v2, v2, v2
+; GFX10-NEXT: flat_store_dword v[0:1], v2
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: load_i16_lo:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-NEXT: v_mov_b32_e32 v2, 0
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: flat_load_d16_b16 v2, v[0:1] offset:8
+; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_u16 v2, v2, v2
+; GFX11-NEXT: flat_store_b32 v[0:1], v2
+; GFX11-NEXT: s_endpgm
%gep = getelementptr inbounds i16, ptr %arg, i32 4
%ld = load i16, ptr %gep, align 2
%vec = insertelement <2 x i16> <i16 poison, i16 0>, i16 %ld, i32 0
@@ -38,10 +130,51 @@ define amdgpu_kernel void @load_i16_lo(ptr %arg, ptr %out) {
ret void
}
-; GCN-LABEL: load_i16_hi:
-; GFX9_11: flat_load_{{short_d16_hi|d16_hi_b16}} v{{[0-9]+}}, v[{{[0-9:]+}}] offset:8{{$}}
-; GFX10: flat_load_short_d16_hi v{{[0-9]+}}, v[{{[0-9:]+}}]{{$}}
define amdgpu_kernel void @load_i16_hi(ptr %arg, ptr %out) {
+; GFX9-LABEL: load_i16_hi:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: v_mov_b32_e32 v2, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v0, s0
+; GFX9-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-NEXT: flat_load_short_d16_hi v2, v[0:1] offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_add_u16 v2, v2, v2
+; GFX9-NEXT: flat_store_dword v[0:1], v2
+; GFX9-NEXT: s_endpgm
+;
+; GFX10-LABEL: load_i16_hi:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX10-NEXT: v_mov_b32_e32 v2, 0
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s0, s0, 8
+; GFX10-NEXT: s_addc_u32 s1, s1, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, s0
+; GFX10-NEXT: v_mov_b32_e32 v1, s1
+; GFX10-NEXT: flat_load_short_d16_hi v2, v[0:1]
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_u16 v2, v2, v2
+; GFX10-NEXT: flat_store_dword v[0:1], v2
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: load_i16_hi:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-NEXT: v_mov_b32_e32 v2, 0
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: flat_load_d16_hi_b16 v2, v[0:1] offset:8
+; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_u16 v2, v2, v2
+; GFX11-NEXT: flat_store_b32 v[0:1], v2
+; GFX11-NEXT: s_endpgm
%gep = getelementptr inbounds i16, ptr %arg, i32 4
%ld = load i16, ptr %gep, align 2
%vec = insertelement <2 x i16> <i16 0, i16 poison>, i16 %ld, i32 1
@@ -50,10 +183,51 @@ define amdgpu_kernel void @load_i16_hi(ptr %arg, ptr %out) {
ret void
}
-; GCN-LABEL: load_half_lo:
-; GFX9_11: flat_load_{{short_d16|d16_b16}} v{{[0-9]+}}, v[{{[0-9:]+}}] offset:8{{$}}
-; GFX10: flat_load_short_d16 v{{[0-9]+}}, v[{{[0-9:]+}}]{{$}}
define amdgpu_kernel void @load_half_lo(ptr %arg, ptr %out) {
+; GFX9-LABEL: load_half_lo:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: v_mov_b32_e32 v2, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v0, s0
+; GFX9-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-NEXT: flat_load_short_d16 v2, v[0:1] offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_add_f16 v2, v2, v2
+; GFX9-NEXT: flat_store_dword v[0:1], v2
+; GFX9-NEXT: s_endpgm
+;
+; GFX10-LABEL: load_half_lo:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX10-NEXT: v_mov_b32_e32 v2, 0
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s0, s0, 8
+; GFX10-NEXT: s_addc_u32 s1, s1, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, s0
+; GFX10-NEXT: v_mov_b32_e32 v1, s1
+; GFX10-NEXT: flat_load_short_d16 v2, v[0:1]
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_f16 v2, v2, v2
+; GFX10-NEXT: flat_store_dword v[0:1], v2
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: load_half_lo:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-NEXT: v_mov_b32_e32 v2, 0
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: flat_load_d16_b16 v2, v[0:1] offset:8
+; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_f16 v2, v2, v2
+; GFX11-NEXT: flat_store_b32 v[0:1], v2
+; GFX11-NEXT: s_endpgm
%gep = getelementptr inbounds half, ptr %arg, i32 4
%ld = load half, ptr %gep, align 2
%vec = insertelement <2 x half> <half poison, half 0xH0000>, half %ld, i32 0
@@ -62,10 +236,51 @@ define amdgpu_kernel void @load_half_lo(ptr %arg, ptr %out) {
ret void
}
-; GCN-LABEL: load_half_hi:
-; GFX9_11: flat_load_{{short_d16_hi|d16_hi_b16}} v{{[0-9]+}}, v[{{[0-9:]+}}] offset:8{{$}}
-; GFX10: flat_load_short_d16_hi v{{[0-9]+}}, v[{{[0-9:]+}}]{{$}}
define amdgpu_kernel void @load_half_hi(ptr %arg, ptr %out) {
+; GFX9-LABEL: load_half_hi:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: v_mov_b32_e32 v2, 0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v0, s0
+; GFX9-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-NEXT: flat_load_short_d16_hi v2, v[0:1] offset:8
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_add_f16 v2, v2, v2
+; GFX9-NEXT: flat_store_dword v[0:1], v2
+; GFX9-NEXT: s_endpgm
+;
+; GFX10-LABEL: load_half_hi:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX10-NEXT: v_mov_b32_e32 v2, 0
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s0, s0, 8
+; GFX10-NEXT: s_addc_u32 s1, s1, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, s0
+; GFX10-NEXT: v_mov_b32_e32 v1, s1
+; GFX10-NEXT: flat_load_short_d16_hi v2, v[0:1]
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_f16 v2, v2, v2
+; GFX10-NEXT: flat_store_dword v[0:1], v2
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: load_half_hi:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-NEXT: v_mov_b32_e32 v2, 0
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: flat_load_d16_hi_b16 v2, v[0:1] offset:8
+; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_f16 v2, v2, v2
+; GFX11-NEXT: flat_store_b32 v[0:1], v2
+; GFX11-NEXT: s_endpgm
%gep = getelementptr inbounds half, ptr %arg, i32 4
%ld = load half, ptr %gep, align 2
%vec = insertelement <2 x half> <half 0xH0000, half poison>, half %ld, i32 1
@@ -74,10 +289,48 @@ define amdgpu_kernel void @load_half_hi(ptr %arg, ptr %out) {
ret void
}
-; GCN-LABEL: load_float_lo:
-; GFX9_11: flat_load_{{dword|b32}} v{{[0-9]+}}, v[{{[0-9:]+}}] offset:16{{$}}
-; GFX10: flat_load_dword v{{[0-9]+}}, v[{{[0-9:]+}}]{{$}}
define amdgpu_kernel void @load_float_lo(ptr %arg, ptr %out) {
+; GFX9-LABEL: load_float_lo:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v0, s0
+; GFX9-NEXT: v_mov_b32_e32 v1, s1
+; GFX9-NEXT: flat_load_dword v2, v[0:1] offset:16
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_mov_b32_e32 v1, s3
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v2, v2, v2
+; GFX9-NEXT: flat_store_dword v[0:1], v2
+; GFX9-NEXT: s_endpgm
+;
+; GFX10-LABEL: load_float_lo:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s0, s0, 16
+; GFX10-NEXT: s_addc_u32 s1, s1, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, s0
+; GFX10-NEXT: v_mov_b32_e32 v1, s1
+; GFX10-NEXT: flat_load_dword v2, v[0:1]
+; GFX10-NEXT: v_mov_b32_e32 v0, s2
+; GFX10-NEXT: v_mov_b32_e32 v1, s3
+; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_f32_e32 v2, v2, v2
+; GFX10-NEXT: flat_store_dword v[0:1], v2
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: load_float_lo:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: flat_load_b32 v2, v[0:1] offset:16
+; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
+; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v2
+; GFX11-NEXT: flat_store_b32 v[0:1], v2
+; GFX11-NEXT: s_endpgm
%gep = getelementptr inbounds float, ptr %arg, i32 4
%ld = load float, ptr %gep, align 4
%v = fadd float %ld, %ld
diff --git a/llvm/test/CodeGen/AMDGPU/lds-misaligned-bug.ll b/llvm/test/CodeGen/AMDGPU/lds-misaligned-bug.ll
index 278ad63b0b76c..7ffc2a6987742 100644
--- a/llvm/...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/137534
More information about the llvm-commits
mailing list